Update README.md

README.md

@@ -18,6 +18,44 @@ IndicGenBench is a multilingual, multi-way parallel benchmark for measuring lang
 
 We extend existing datasets in Cross-lingual Summarization (CrossSum), Machine Translation (FLORES), Multi-lingual Question Answering (XQuAD), and Cross-lingual Question Answering (XorQA) by collecting human translations for English examples into the target Indic languages. Please see the paper for details.
 
+### Dataset Structure
+
+#### Dataset Fields
+
+- text: The paragraph which is to be summarized. This paragraph is in English.
+- summary: summary of the paragraph in the langauge specified by the "lang" field.
+- lang: language code of the targeted language. 'summary' is in this language.
+
+#### Dataset Instances
+
+```
+{
+ 'lang': 'as',
+ 'source_url': 'https://www.bbc.com/news/technology-39193008',
+ 'summary': 'Wikileaksএ এনে কিছু সবিশেষ প্ৰকাশ কৰিছে, যিবোৰ ইয়াৰ মতে এইবোৰ '
+            'হৈছে চিআইএৰ দ্বাৰা ব্যৱহৃত বিস্তৃত পৰিসৰৰ হেকিং সঁজুলি।',
+ 'target_url': 'https://www.bbc.com/news/technology-39193008',
+ 'text': 'By Leo KelionTechnology desk editor The alleged cyber-weapons are '
+         'said to include malware that targets Windows, Android, iOS, OSX and '
+         'Linux computers as well as internet routers. Some of the software is '
+         "Reported [...TRUNCATED…].'
+}
+```
+
+#### Dataset Usage Instructions
+
+for loading the dataset and printing the first 5 examples from the validation set:
+
+```
+dataset = load_dataset('google/IndicGenBench_crosssum_in', field="examples")
+
+for i, batch in enumerate(dataset['validation']):
+    pprint(batch)
+    if i==4:
+        break
+```
+
+
 ### Dataset Description
 
 <!-- Provide a longer summary of what this dataset is. -->